Associated manuscript: A guide to calibration plots for clinical prediction models predicting survival outcomes
This document contains supplementary Figures associated with the aforementioned manuscript. The supplementary material is split into 7 sections:
We present the calibration plots under the four different censoring mechanisms when the coefficient for each covariate is 0.5 as opposed to 0.3 (0.2 instead of 0.12 for the cubic term). Compared to the main analysis, we see increased magnitude in the bias when a variable predictive of both the outcome and censoring mechanism is omitted from the clinical prediction model (e.g. see Figure S3, compared to Figure 3 from the manuscript).
Figure S1: Calibration plots under C-DGM-R, coefficients of 0.5/0.2
—————————————————————————————————————————————————————————
Figure S2: Calibration plots under C-DGM-SL, coefficients of 0.5/0.2
—————————————————————————————————————————————————————————
Figure S3: Calibration plots under C-DGM-ML, coefficients of 0.5/0.2
—————————————————————————————————————————————————————————
Figure S4: Calibration plots under C-DGM-QU, coefficients of 0.5/0.2
We see that the performance of each calibration method remains unchanged from the main analysis, however the calibration of the clinical prediction model itself is impcated. Specifically, the model is poorly calibrated if a covariate is omitted which is predictive of both the outcome and the censoring mechanism in the development dataset. This calibration of the model is worse than when these variables were omitted, but censoring was not applied in the development dataset (see main analysis). The coefficients in these plots have been returned to 0.3/0.12, the same as those from the main analysis.
Figure S1: Calibration plots under C-DGM-R, censoring applied to development and validation datasets
—————————————————————————————————————————————————————————
Figure S2: Calibration plots under C-DGM-SL, censoring applied to development and validation datasets
—————————————————————————————————————————————————————————
Figure S3: Calibration plots under C-DGM-ML, censoring applied to development and validation datasets
—————————————————————————————————————————————————————————
Figure S4: Calibration plots under C-DGM-QU, censoring applied to development and validation datasets